crop = x[start:start + self.dim[0], :]
X[i, ] = crop
return X
########################################################################
########################################################################
# main 00_train.py
########################################################################
if __name__ == "__main__":
# check mode
# "development": mode == True
# "evaluation": mode == False
mode, target = com.command_line_chk()
if mode is None:
sys.exit(-1)
# make output directory
os.makedirs(param["model_directory"], exist_ok=True)
# initialize the visualizer
visualizer = visualizer()
# load base_directory list
dirs = com.select_dirs(param=param, mode=mode, target=target)
# loop of the base directory (machine types)
for idx, target_dir in enumerate(dirs):
print("\n===========================")
com.logger.exception("no_wav_file!!")
com.logger.info("=========================================")
return files, labels
########################################################################
########################################################################
# main 01_test.py
########################################################################
if __name__ == "__main__":
# check mode
# "development": mode == True
# "evaluation": mode == False
mode = com.command_line_chk()
if mode is None:
sys.exit(-1)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
# make output result directory
os.makedirs(PARAM["result_directory"], exist_ok=True)
# load base directory
dirs = com.select_dirs(param=PARAM, mode=mode)
# initialize lines in csv for AUC and pAUC
csv_lines = []
# loop of the base directory
def main():
"""
perform model training and validation
"""
# check mode
# "development": mode == True
# "evaluation": mode == False
mode = com.command_line_chk() # constant: True or False
if mode is None:
sys.exit(-1)
# make output directory
os.makedirs(PARAM["model_directory"], exist_ok=True)
# load base_directory list
dir_list = com.select_dirs(param=PARAM, mode=mode)
# loop of the base directory (for each machine)
dir_list = [
'/work/tamamori/dcase2020/dcase2020_task2_baseline/dev_data/ToyCar'
]
for idx, target_dir in enumerate(dir_list):
com.logger.info("===========================")
com.logger.info("[%d/%d] %s", idx + 1, len(dir_list), target_dir)
com.logger.info("============== DATASET_GENERATOR ==============")
dcase_dataset = DcaseDataset(target_dir)
n_samples = len(dcase_dataset) # total number of frames
train_size = int(n_samples *
(1.0 - PARAM["training"]["validation_split"]))
dataset = {"train": None, "val": None}
dataset["train"] = Subset(dcase_dataset, list(range(0, train_size)))
dataset["val"] = Subset(dcase_dataset,
list(range(train_size, n_samples)))
com.logger.info("============== DATALOADER_GENERATOR ==============")
data_loader = {"train": None, "val": None}
data_loader["train"] = torch.utils.data.DataLoader(
dataset["train"],
batch_size=PARAM["training"]["batch_size"],
shuffle=PARAM["training"]["shuffle"],
drop_last=True)
data_loader["val"] = torch.utils.data.DataLoader(
dataset["val"],
batch_size=PARAM["training"]["batch_size"],
shuffle=False,
drop_last=False)
com.logger.info("============== MODEL TRAINING ==============")
model = VAE(x_dim=PARAM["feature"]["n_mels"] *
PARAM["feature"]["frames"],
h_dim=PARAM["model"]["hidden_dim"],
z_dim=PARAM["model"]["latent_dim"],
n_hidden=PARAM["model"]["n_hidden"]).to(DEVICE)
optimizer = optim.Adam(model.parameters(),
weight_decay=PARAM["training"]["weight_decay"])
criterion = nn.MSELoss(reduction='mean')
summary(model,
input_size=(PARAM["feature"]["n_mels"] *
PARAM["feature"]["frames"], ))
loss = {"train": 0.0, "val": 0.0}
for epoch in range(1, PARAM["training"]["epochs"] + 1):
loss["train"] = training(model, data_loader["train"], optimizer,
criterion)
loss["val"] = validation(model, data_loader["train"], criterion)
com.logger.info(
"Epoch %2d: Average train_loss: %.6f, "
"Average validation_loss: %.6f", epoch, loss["train"],
loss["val"])
com.logger.info("============== SAVE MODEL ==============")
torch.save(
model.state_dict(), "%s/model_ae_%s.pt" %
(PARAM["model_directory"], os.path.split(target_dir)[1]))
com.logger.info(
"save_model -> %s", "%s/model_ae_%s.pt" %
(PARAM["model_directory"], os.path.split(target_dir)[1]))
com.logger.info("============== END TRAINING ==============")
